[llvm-dev] 'git llvm push' not working for me on Windows

Fri Feb 1 18:37:16 PST 2019

I already figured it out. Turned out some sources
like clang/lib/Basic/Version.cpp contain Subversion keywords (in this
particular case it's $URL$). Subversion would automatically substitute
these in the internal checkout that's used by git-llvm-push, but Git
wouldn't, so when trying to apply the patch generated by Git you'll get and
error because while Git patch has $URL$, the Subversion checkout would have
something like $URL: https://llvm.org/svn/llvm-project/cfe/trunk/$. The
workaround I came up with is to do the keyword substitution in the
generated patch (
https://github.com/llvm/llvm-project/blob/master/llvm/utils/git-svn/git-llvm#L321
).

On Fri, Feb 1, 2019 at 2:56 PM <paul.robinson at sony.com> wrote:

> The usual workaround is to do `rm –rf .git/llvm-upstream-svn` and try
> again.
>
> You could also patch llvm/utils/git-svn/git-llvm to add the `--verbose`
> option to the `git apply` command, which would provide better diagnostic
> output.
>
> HTH,
>
> --paulr
>
>
>
> *From:* Petr Hosek [mailto:phosek at chromium.org]
> *Sent:* Wednesday, January 30, 2019 10:52 PM
> *To:* Robinson, Paul
> *Cc:* James Y Knight; llvm-dev; vedant_kumar at apple.com
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> I just hit this failure as well. I've uploaded the log to
> https://reviews.llvm.org/P8126. The error message is:
>
>
>
> `git apply -p2 -` returned 1
>
> error: patch failed: lib/Basic/Version.cpp:14
>
> error: lib/Basic/Version.cpp: patch does not apply
>
> Patch doesn't apply: maybe you should try `git pull -r` first?
>
>
>
> On Thu, Nov 29, 2018 at 1:18 PM via llvm-dev <llvm-dev at lists.llvm.org>
> wrote:
>
> Excellent, thanks!
>
> --paulr
>
>
>
> *From:* James Y Knight [mailto:jyknight at google.com]
> *Sent:* Thursday, November 29, 2018 12:00 PM
> *To:* Robinson, Paul
> *Cc:* Mikhail Ramalho; vedant_kumar at apple.com; llvm-dev
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> Aha! From your output I figured out what I screwed up. Sorry about the
> trouble...
>
>
>
> Illustration of the issue:
>
> svn checkout --depth=empty https://llvm.org/svn/llvm-project
>
> svn update --depth=immediates --parents llvm/trunk/test/DebugInfo
>
> svn update --depth=immediates --parents llvm/trunk/test/DebugInfo/Generic
>
>
>
> After these commands, I expected Generic/ to be populated, but it is not,
> because the "--depth=immediates" *creates* the directory, but then it marks
> it as depth=empty. And then, later, calling svn update --depth=immediates
> _doesn't override that_. This is apparently not a bug in subversion, that's
> is how it is "supposed" to work. It's just super confusing. If I use
> --depth=files instead of --depth=immediates, that fixes this issue, since
> in that case, it won't create the subdirs and mark them as depth=empty.
>
>
>
> Also, the --parents flag causes the same issue for parent directories.
> Ugh. After the following, I'd expected llvm/trunk/ to get populated, but it
> doesn't get populated, because the first 'svn update' created it with
> sticky depth=empty:
>
> svn checkout --depth=empty https://llvm.org/svn/llvm-project
>
> svn update --depth=files --parents llvm/trunk/test/DebugInfo
>
> svn update --depth=files --parents llvm/trunk/
>
>
>
> Subversion has a --set-depth= argument which "fixes" this, but
> unfortunately, that argument *also* recursively deletes any subdirs which
> have already been populated. (I just want to _add_ things, not delete
> already-downloaded data.)
>
>
>
> In summary: subversion's sparse-checkout command-line UI is really
> confusing and annoying! Ugh!
>
>
>
> I've pushed a commit (r347883) to hopefully fix this issue. This time for
> sure.
>
>
>
> You will probably need to remove the .git/llvm-upstream-svn directory, one
> last time, for the fix to work.
>
>
>
> On Wed, Nov 28, 2018 at 3:55 PM <paul.robinson at sony.com> wrote:
>
> Poking around in the .git\llvm-upstream-svn tree, I find that
> llvm\trunk\test\DebugInfo\Generic is empty, as are all the other
> subdirectories of test\DebugInfo that I tried.  I have other files in the
> checkin that are in leaf directories but those files all exist.
>
> I hacked git-llvm to add a –verbose option:
>
>
>
> `git apply --verbose -p2 -` returned 1
>
> Checking patch include/llvm/IR/DebugInfoFlags.def...
>
> Checking patch include/llvm/IR/DebugInfoMetadata.h...
>
> Checking patch lib/AsmParser/LLLexer.cpp...
>
> Checking patch lib/AsmParser/LLParser.cpp...
>
> Checking patch lib/AsmParser/LLToken.h...
>
> Checking patch lib/Bitcode/Reader/MetadataLoader.cpp...
>
> Checking patch lib/Bitcode/Writer/BitcodeWriter.cpp...
>
> Checking patch lib/IR/AsmWriter.cpp...
>
> Checking patch lib/IR/DebugInfoMetadata.cpp...
>
> Checking patch test/Assembler/disubprogram.ll...
>
> Checking patch test/Assembler/invalid-disubprogram-uniqued-definition.ll...
>
> Checking patch test/Bindings/llvm-c/debug_info.ll...
>
> Checking patch test/Bitcode/DISubprogram-distinct-definitions.ll...
>
> Checking patch test/Bitcode/DISubprogram-v4.ll...
>
> Checking patch test/Bitcode/DISubprogram-v4.ll.bc...
>
> Checking patch test/DebugInfo/Generic/invalid.ll...
>
> error: test/DebugInfo/Generic/invalid.ll: No such file or directory
>
> Checking patch test/DebugInfo/debugify.ll...
>
> Checking patch test/Linker/replaced-function-matches-first-subprogram.ll...
>
> Checking patch test/Transforms/GCOVProfiling/three-element-mdnode.ll...
>
> Patch doesn't apply: maybe you should try `git pull -r` first?
>
>
>
>
>
> *From:* llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] *On Behalf Of *via
> llvm-dev
> *Sent:* Wednesday, November 28, 2018 2:39 PM
> *To:* mikhail.ramalho at gmail.com; jyknight at google.com
> *Cc:* llvm-dev at lists.llvm.org; vedant_kumar at apple.com
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> And now it's failing for me also… complaining about "no such file or
> directory" on a file that plainly exists.
>
> Verbose log attached.
>
> --paulr
>
>
>
> *From:* Mikhail Ramalho [mailto:mikhail.ramalho at gmail.com]
> *Sent:* Wednesday, November 28, 2018 12:27 PM
> *To:* jyknight at google.com
> *Cc:* Robinson, Paul; llvm-dev at lists.llvm.org; vedant_kumar at apple.com
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> OK, I've managed to do it:
>
>
>
> I was trying to push it from a build/ directory, maybe that's why the git
> apply was failing.
>
>
>
> Pushing the commit from the root of the repo worked.
>
>
>
> Em qua, 28 de nov de 2018 às 16:40, Mikhail Ramalho <
> mikhail.ramalho at gmail.com> escreveu:
>
> Hi,
>
>
>
> The patch only changes one file in clang. Here's the patch:
> https://reviews.llvm.org/D54974
>
>
>
> Attached the log. It seems to complain about the git apply -p2.
>
>
>
> Maybe it's something related to arc?
>
>
>
> Em qua, 28 de nov de 2018 às 15:37, James Y Knight <jyknight at google.com>
> escreveu:
>
> Can you please run "git llvm --verbose push" and send me the output?
>
>
>
> BTW, I've just committed a fix for the handling of binary patch data when
> you're running it under python3.X. I don't believe that was a regression
> from my recent changes, but if you're using python3, you could see if that
> fixes it for you.
>
>
>
> On Tue, Nov 27, 2018 at 7:20 PM Mikhail Ramalho via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> I'm facing this now on Linux.
>
>
>
> Any idea how to fix it?
>
>
>
> Em seg, 26 de nov de 2018 às 20:22, via llvm-dev <llvm-dev at lists.llvm.org>
> escreveu:
>
> I've verified that none of the files I tried to check in had Windows-style
> line endings. It's something else.
>
> --paulr
>
>
>
> *From:* vsk at apple.com [mailto:vsk at apple.com] *On Behalf Of *Vedant Kumar
> *Sent:* Thursday, November 22, 2018 9:38 AM
> *To:* Robinson, Paul
> *Cc:* zturner at google.com; jyknight at google.com; llvm-dev at lists.llvm.org
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> As a test case, try committing a change to
> clang/test/SemaCXX/sourceranges.cpp?
>
>
>
> I believe most editors force Windows-style line endings for that file
> (they show up in my vim as “^M”). I recently tried to commit a change to
> that file using “git llvm push” on macOS, but hit what I think is the same
> issue. At least, the error message was the same.
>
> vedant (sent from my iPhone)
>
>
> On Nov 21, 2018, at 10:06 AM, via llvm-dev <llvm-dev at lists.llvm.org>
> wrote:
>
> FTR the commit where it failed for me had both llvm and clang changes.  I
> wonder if that contributed.
>
> --paulr
>
>
>
> *From:* llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org
> <llvm-dev-bounces at lists.llvm.org>] *On Behalf Of *via llvm-dev
> *Sent:* Monday, November 19, 2018 6:01 PM
> *To:* zturner at google.com; jyknight at google.com
> *Cc:* llvm-dev at lists.llvm.org
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> Hm.  Just now it worked for me for r347271.  If it happens again I'll try
> verbose mode and attach a log.
>
> Thanks,
>
> --paulr
>
>
>
> *From:* Zachary Turner [mailto:zturner at google.com <zturner at google.com>]
> *Sent:* Monday, November 19, 2018 5:26 PM
> *To:* James Y Knight
> *Cc:* Robinson, Paul; llvm-dev
> *Subject:* Re: [llvm-dev] 'git llvm push' not working for me on Windows
>
>
>
> Usually every time I've seen that error message, it's been related to line
> ending normalization.  But James is right, I did use it successfully this
> morning as well as yesterday.
>
>
>
> On Mon, Nov 19, 2018 at 1:03 PM James Y Knight via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> It does sound like I must've broken something, but I believe zturner used
> it on windows successfully. Perhaps you could help me debug the issue?
> Possibly running it with --verbose would show something useful.
>
>
>
> On Mon, Nov 19, 2018 at 1:32 PM <paul.robinson at sony.com> wrote:
>
> I am consistently getting:
>
> `git apply -p2 -` returned 1
> error: include/llvm/IR/DIBuilder.h: No such file or directory
> [[ etc ]]
> Patch doesn't apply; maybe you should try `git pull -r` first?
>
> My usual response to a problem from git-llvm (which is most often
> an anti-virus issue) is to blow away .git\llvm-upstream-svn but
> that doesn't help this time.
>
> I see James Knight did a "performance improvement" on Friday,
> perhaps that doesn't work so well on Windows?  The main
> performance cost is one-time, and as long as my anti-virus isn't
> trashing things behind my back the performance is just fine;
> so it's not clear this is really a necessary improvement.
> If it's interfering with developing the new git repo, then I'd
> suggest making it optional or platform-dependent.
>
> As a workaround I fetched the previous version of git-llvm
> and put it earlier in my PATH, that worked.
>
> --paulr
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
>
>
> --
>
>
>
> Mikhail Ramalho.
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
>
>
> --
>
>
>
> Mikhail Ramalho.
>
>
>
>
> --
>
>
>
> Mikhail Ramalho.
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20190201/ad1639b6/attachment.html>